1

I have seen worked examples of bootstrapping coefficient or odds ratio of logistic regression. As logistic regression does not assume any distributional assumption, what is the purpose of bootstrapping? Does it help to check if the model is overfitted by comparing the original confidence interval of the coefficients to the bootstrapped ones? In the event where the purpose of the regression is inference or descriptive does bootstrapping have any use at all?

I understand that it helps to provide level of uncertainty of the coefficient when the sample is small. But how about when sample is not small (>50,000)? Why would the usual confidence interval be not sufficient? Are there other purposes that are more related to model fitting strategies?

Appreciate if anyone can provide some references on how bootstrapping helps.

tatami
  • 695
  • 1
  • 7
  • 24
  • 3
    Possible duplicate of [bootstrap method](https://stats.stackexchange.com/questions/295328/bootstrap-method) – Tim Aug 10 '17 at 12:09
  • 1
    You say *[...] does not assume any distributional assumption [...].* There are distributional assumptions involved when computing [CI](https://en.wikipedia.org/wiki/Confidence_interval)s. At least about its bounds. Which is not the case of those computed over the bootstrap distribution of your coefficient. – keepAlive Aug 10 '17 at 16:00
  • https://stats.stackexchange.com/questions/65244/how-to-determine-the-accuracy-of-logistic-regression-in-r – kjetil b halvorsen Feb 21 '20 at 03:51

0 Answers0